← back
The State of MCP observability: Observable.tools — Alex Volkov and Benjamin Eckel, W&B and Dylibso
Takeaway
Treat MCP-powered agents as a distributed system and propagate OpenTelemetry context through MCP's meta payload to recover end-to-end tracing.
Summary
- W&B (Weave) and Dylibso (MCP.run) argue MCP creates an observability blind spot once agents go from owning all code to invoking remote MCP servers.
- Weave now supports MCP via env var MCP_TRACE_LIST_OPERATION — works for Python client/server you own, but stops at third-party MCP servers.
- Pitches Observable.tools: a vendor-neutral manifesto using OpenTelemetry (traces, spans, sinks) with context propagation through MCP's meta payload to stitch client+server traces.
- Same-domain vs different-domain MCP servers behave differently — third-party servers appear as black-box spans without standardized OTel propagation.
mcpobservabilityopentelemetry
Original description
AI Engineers deserve observable tools! MCP getting adoption means that less and less of your agents code is running under your control, and this has DX and observability challenges, let's fix that! Join Alex Volkov from Weights & Biases and Steve Manual from mcp.run on this recap of the current state of MCP observability, including the observable.tools initiative, a recap of where the field stands and what to look forward to + a practical example of MCP tool usage evaluation framework from mcp.run! About Alex Volkov Alex Volkov is an AI Evangelist at Weights & Biases as well as the founder and host of ThursdAI, a weekly newsletter and podcast that explores the latest innovations in AI, their practical applications, and the open-source AI community. Alex is an AI startup founder with 20 years of full-stack software engineering experience, offering a deep well of insights into AI innovation. He’s celebrated for his ability to clarify and summarize the complexities of the rapid AI advances and advocating for its beneficial uses. About Benjamin Eckel Benjamin has over a decade of experience as a software engineer and is the co-founder and CTO of Dylibso. He previously led DX at Recurly and worked on integrations and edge observability at Datadog. Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter